Classification¶

A detailed walkthrough of DeepGBoostClassifier covering:

1. Binary classification — Breast Cancer dataset
2. Multiclass classification — Iris dataset
3. predict_proba and calibration
4. ROC-AUC and confusion matrix
5. Scikit-learn pipeline integration
6. Feature importances
7. String labels

In [1]:

import numpy as np
import matplotlib.pyplot as plt
from sklearn.datasets import load_breast_cancer, load_iris
from sklearn.model_selection import train_test_split, cross_val_score
from sklearn.metrics import (
    accuracy_score,
    roc_auc_score,
    ConfusionMatrixDisplay,
    RocCurveDisplay,
)
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
from sklearn.base import clone

from deepgboost import DeepGBoostClassifier, plot_importance

import numpy as np import matplotlib.pyplot as plt from sklearn.datasets import load_breast_cancer, load_iris from sklearn.model_selection import train_test_split, cross_val_score from sklearn.metrics import ( accuracy_score, roc_auc_score, ConfusionMatrixDisplay, RocCurveDisplay, ) from sklearn.pipeline import Pipeline from sklearn.preprocessing import StandardScaler from sklearn.base import clone from deepgboost import DeepGBoostClassifier, plot_importance

---

1. Binary Classification — Breast Cancer¶

In [2]:

bc = load_breast_cancer()
X, y = bc.data, bc.target
X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.2, random_state=42
)

clf = DeepGBoostClassifier(
    n_trees=5,
    n_layers=10,
    max_depth=4,
    learning_rate=0.15,
    random_state=42,
)
clf.fit(X_train, y_train)

y_pred = clf.predict(X_test)
y_proba = clf.predict_proba(X_test)

acc = accuracy_score(y_test, y_pred)
auc = roc_auc_score(y_test, y_proba[:, 1])

print(f"Accuracy: {acc:.4f}")
print(f"ROC-AUC:  {auc:.4f}")
print(f"Classes:  {clf.classes_}")
print(f"n_classes: {clf.n_classes_}")

bc = load_breast_cancer() X, y = bc.data, bc.target X_train, X_test, y_train, y_test = train_test_split( X, y, test_size=0.2, random_state=42 ) clf = DeepGBoostClassifier( n_trees=5, n_layers=10, max_depth=4, learning_rate=0.15, random_state=42, ) clf.fit(X_train, y_train) y_pred = clf.predict(X_test) y_proba = clf.predict_proba(X_test) acc = accuracy_score(y_test, y_pred) auc = roc_auc_score(y_test, y_proba[:, 1]) print(f"Accuracy: {acc:.4f}") print(f"ROC-AUC: {auc:.4f}") print(f"Classes: {clf.classes_}") print(f"n_classes: {clf.n_classes_}")

Accuracy: 0.9649
ROC-AUC:  0.9971
Classes:  [0 1]
n_classes: 2

ROC Curve¶

In [3]:

fig, ax = plt.subplots(figsize=(5, 4))
RocCurveDisplay.from_predictions(
    y_test, y_proba[:, 1], ax=ax, name=f"DeepGBoost (AUC={auc:.3f})"
)
ax.set_title("Breast Cancer — ROC Curve")
plt.tight_layout()
plt.show()

fig, ax = plt.subplots(figsize=(5, 4)) RocCurveDisplay.from_predictions( y_test, y_proba[:, 1], ax=ax, name=f"DeepGBoost (AUC={auc:.3f})" ) ax.set_title("Breast Cancer — ROC Curve") plt.tight_layout() plt.show()

Confusion Matrix¶

In [4]:

fig, ax = plt.subplots(figsize=(4, 3))
ConfusionMatrixDisplay.from_predictions(
    y_test, y_pred, display_labels=bc.target_names, ax=ax
)
ax.set_title("Breast Cancer — Confusion Matrix")
plt.tight_layout()
plt.show()

fig, ax = plt.subplots(figsize=(4, 3)) ConfusionMatrixDisplay.from_predictions( y_test, y_pred, display_labels=bc.target_names, ax=ax ) ax.set_title("Breast Cancer — Confusion Matrix") plt.tight_layout() plt.show()

Predicted probabilities¶

In [5]:

fig, axes = plt.subplots(1, 2, figsize=(9, 3), sharey=True)
for cls_idx, (ax, name) in enumerate(zip(axes, bc.target_names)):
    mask = y_test == cls_idx
    ax.hist(y_proba[mask, 1], bins=20, edgecolor="k", linewidth=0.4)
    ax.set_title(f"P(malignant | true={name})")
    ax.set_xlabel("Predicted probability")
    ax.set_ylabel("Count")
plt.suptitle("Probability distributions by true class", y=1.02)
plt.tight_layout()
plt.show()

fig, axes = plt.subplots(1, 2, figsize=(9, 3), sharey=True) for cls_idx, (ax, name) in enumerate(zip(axes, bc.target_names)): mask = y_test == cls_idx ax.hist(y_proba[mask, 1], bins=20, edgecolor="k", linewidth=0.4) ax.set_title(f"P(malignant | true={name})") ax.set_xlabel("Predicted probability") ax.set_ylabel("Count") plt.suptitle("Probability distributions by true class", y=1.02) plt.tight_layout() plt.show()

---

2. Multiclass Classification — Iris¶

In [6]:

iris = load_iris()
Xi, yi = iris.data, iris.target
Xi_train, Xi_test, yi_train, yi_test = train_test_split(
    Xi, yi, test_size=0.2, random_state=42
)

clf_mc = DeepGBoostClassifier(
    n_trees=5,
    n_layers=8,
    max_depth=3,
    learning_rate=0.1,
    random_state=0,
)
clf_mc.fit(Xi_train, yi_train)

yi_pred = clf_mc.predict(Xi_test)
yi_proba = clf_mc.predict_proba(Xi_test)

acc_mc = accuracy_score(yi_test, yi_pred)
print(f"Accuracy: {acc_mc:.4f}")
print(f"Classes:  {clf_mc.classes_}")
print(
    f"predict_proba shape: {yi_proba.shape}  (rows sum to 1: {np.allclose(yi_proba.sum(1), 1)})"
)

iris = load_iris() Xi, yi = iris.data, iris.target Xi_train, Xi_test, yi_train, yi_test = train_test_split( Xi, yi, test_size=0.2, random_state=42 ) clf_mc = DeepGBoostClassifier( n_trees=5, n_layers=8, max_depth=3, learning_rate=0.1, random_state=0, ) clf_mc.fit(Xi_train, yi_train) yi_pred = clf_mc.predict(Xi_test) yi_proba = clf_mc.predict_proba(Xi_test) acc_mc = accuracy_score(yi_test, yi_pred) print(f"Accuracy: {acc_mc:.4f}") print(f"Classes: {clf_mc.classes_}") print( f"predict_proba shape: {yi_proba.shape} (rows sum to 1: {np.allclose(yi_proba.sum(1), 1)})" )

Accuracy: 1.0000
Classes:  [0 1 2]
predict_proba shape: (30, 3)  (rows sum to 1: True)

In [7]:

fig, ax = plt.subplots(figsize=(4, 3))
ConfusionMatrixDisplay.from_predictions(
    yi_test, yi_pred, display_labels=iris.target_names, ax=ax
)
ax.set_title(f"Iris — Confusion Matrix (acc={acc_mc:.3f})")
plt.tight_layout()
plt.show()

fig, ax = plt.subplots(figsize=(4, 3)) ConfusionMatrixDisplay.from_predictions( yi_test, yi_pred, display_labels=iris.target_names, ax=ax ) ax.set_title(f"Iris — Confusion Matrix (acc={acc_mc:.3f})") plt.tight_layout() plt.show()

Probability simplex (2-D PCA projection)¶

In [8]:

from sklearn.decomposition import PCA

pca = PCA(n_components=2).fit(Xi)
Xi_2d = pca.transform(Xi_test)

fig, axes = plt.subplots(1, 3, figsize=(12, 3))
colors = ["#1f77b4", "#ff7f0e", "#2ca02c"]
for k, (ax, species) in enumerate(zip(axes, iris.target_names)):
    sc = ax.scatter(
        Xi_2d[:, 0],
        Xi_2d[:, 1],
        c=yi_proba[:, k],
        cmap="RdYlGn",
        vmin=0,
        vmax=1,
        edgecolors="k",
        linewidths=0.3,
        s=40,
    )
    ax.set_title(f"P({species})")
    ax.set_xlabel("PC1")
    ax.set_ylabel("PC2")
    plt.colorbar(sc, ax=ax)
plt.suptitle("Predicted class probabilities (PCA 2-D)", y=1.02)
plt.tight_layout()
plt.show()

from sklearn.decomposition import PCA pca = PCA(n_components=2).fit(Xi) Xi_2d = pca.transform(Xi_test) fig, axes = plt.subplots(1, 3, figsize=(12, 3)) colors = ["#1f77b4", "#ff7f0e", "#2ca02c"] for k, (ax, species) in enumerate(zip(axes, iris.target_names)): sc = ax.scatter( Xi_2d[:, 0], Xi_2d[:, 1], c=yi_proba[:, k], cmap="RdYlGn", vmin=0, vmax=1, edgecolors="k", linewidths=0.3, s=40, ) ax.set_title(f"P({species})") ax.set_xlabel("PC1") ax.set_ylabel("PC2") plt.colorbar(sc, ax=ax) plt.suptitle("Predicted class probabilities (PCA 2-D)", y=1.02) plt.tight_layout() plt.show()

---

3. Scikit-learn Pipeline¶

In [9]:

pipe = Pipeline(
    [
        ("scaler", StandardScaler()),
        ("clf", DeepGBoostClassifier(n_trees=5, n_layers=8, random_state=0)),
    ]
)

cv_scores = cross_val_score(pipe, Xi, yi, cv=5, scoring="accuracy")
print(f"5-fold CV accuracy: {cv_scores.mean():.4f} ± {cv_scores.std():.4f}")

pipe = Pipeline( [ ("scaler", StandardScaler()), ("clf", DeepGBoostClassifier(n_trees=5, n_layers=8, random_state=0)), ] ) cv_scores = cross_val_score(pipe, Xi, yi, cv=5, scoring="accuracy") print(f"5-fold CV accuracy: {cv_scores.mean():.4f} ± {cv_scores.std():.4f}")

5-fold CV accuracy: 0.9133 ± 0.0980

---

4. Feature Importances¶

In [10]:

# Breast Cancer — binary
fig, ax = plot_importance(
    clf,
    feature_names=list(bc.feature_names),
    max_features=10,
    title="Breast Cancer — Top 10 Feature Importances",
)
plt.tight_layout()
plt.show()

# Breast Cancer — binary fig, ax = plot_importance( clf, feature_names=list(bc.feature_names), max_features=10, title="Breast Cancer — Top 10 Feature Importances", ) plt.tight_layout() plt.show()

In [11]:

# Iris — multiclass (importances averaged across OvR models)
fig, ax = plot_importance(
    clf_mc,
    feature_names=list(iris.feature_names),
    title="Iris — Feature Importances (OvR average)",
)
plt.tight_layout()
plt.show()

# Iris — multiclass (importances averaged across OvR models) fig, ax = plot_importance( clf_mc, feature_names=list(iris.feature_names), title="Iris — Feature Importances (OvR average)", ) plt.tight_layout() plt.show()

---

5. String Labels¶

In [12]:

# DeepGBoostClassifier handles arbitrary label types via LabelEncoder internally
y_str_train = np.array(["setosa", "versicolor", "virginica"])[yi_train]
y_str_test = np.array(["setosa", "versicolor", "virginica"])[yi_test]

clf_str = DeepGBoostClassifier(n_trees=5, n_layers=8, random_state=0)
clf_str.fit(Xi_train, y_str_train)

preds_str = clf_str.predict(Xi_test)
print("Predicted labels (sample):", preds_str[:10])
print("Classes stored:", clf_str.classes_)
print(f"Accuracy: {accuracy_score(y_str_test, preds_str):.4f}")

# DeepGBoostClassifier handles arbitrary label types via LabelEncoder internally y_str_train = np.array(["setosa", "versicolor", "virginica"])[yi_train] y_str_test = np.array(["setosa", "versicolor", "virginica"])[yi_test] clf_str = DeepGBoostClassifier(n_trees=5, n_layers=8, random_state=0) clf_str.fit(Xi_train, y_str_train) preds_str = clf_str.predict(Xi_test) print("Predicted labels (sample):", preds_str[:10]) print("Classes stored:", clf_str.classes_) print(f"Accuracy: {accuracy_score(y_str_test, preds_str):.4f}")

Predicted labels (sample): ['versicolor' 'setosa' 'virginica' 'versicolor' 'versicolor' 'setosa'
 'versicolor' 'virginica' 'versicolor' 'versicolor']
Classes stored: ['setosa' 'versicolor' 'virginica']
Accuracy: 1.0000

---

6. Sklearn get_params / set_params / clone¶

In [13]:

clf_base = DeepGBoostClassifier(
    n_trees=5, n_layers=8, learning_rate=0.1, random_state=0
)
print("get_params:", clf_base.get_params())

clf_base.set_params(n_layers=12, learning_rate=0.05)
print(
    "After set_params — n_layers:",
    clf_base.n_layers,
    "  learning_rate:",
    clf_base.learning_rate,
)

clf_cloned = clone(clf_base)
print("Cloned (no fitted attrs):", not hasattr(clf_cloned, "_binary_model"))

clf_base = DeepGBoostClassifier( n_trees=5, n_layers=8, learning_rate=0.1, random_state=0 ) print("get_params:", clf_base.get_params()) clf_base.set_params(n_layers=12, learning_rate=0.05) print( "After set_params — n_layers:", clf_base.n_layers, " learning_rate:", clf_base.learning_rate, ) clf_cloned = clone(clf_base) print("Cloned (no fitted attrs):", not hasattr(clf_cloned, "_binary_model"))

get_params: {'early_stopping_rounds': None, 'eval_metric': None, 'hessian_reg': 0.0, 'learning_rate': 0.1, 'linear_alpha': 1.0, 'linear_projection': False, 'max_depth': None, 'max_features': None, 'n_jobs': 1, 'n_layers': 8, 'n_trees': 5, 'objective': None, 'random_state': 0, 'subsample_min_frac': 0.3, 'weight_solver': 'nnls'}
After set_params — n_layers: 12   learning_rate: 0.05
Cloned (no fitted attrs): True